Analysis of Purchase History Data Based on a New Latent Class Model for RFM Analysis

Abstract

Recently, it has become easier to make use of various kinds of information on customers (e.g. customers’ purchase history), due to the development of information technology. Especially in the marketing field, in fact, many companies try to employ customer segmentation for the services customization which leads to increase customer loyalty and to keep high customer retention. One of the well-known approaches for the customer analysis based on purchase history data is the RFM analysis. The RFM analysis is usually used to segment customers into several groups by using three variables; how long it has been since their last purchase, how many times they purchased, and how much they spent. However, the conventional method of the RFM analysis did not assume a generative model. Therefore, when applying to an actual data set and scoring each index of R, F, M scores, several problems occur. The main problem is that an analyst should arbitrarily decide the threshold for the scores of RFM. On the other hand, in the field of machine learning, the probabilistic latent semantic analysis is widely used for soft clustering. The latent class model enables us to cluster customers into latent classes and to calculate the assignment probabilities of each customer to each latent class. In this paper, we propose a new latent class model for the RFM analysis based on the purchase history data. The proposed model enables to decide the scoring of RFM and segment customers automatically, and the soft clustering approach helps the interpretation of the result. Furthermore, the proposed model takes account of the generation model of RFM scores. From the result of actual data analysis, it became clear that it is possible to extract latent classes that express the statistical characteristics of data well. Given a generative model estimated from the given data, it is also possible to predict future purchase behaviors of customers or to generate virtual data for simulation analysis and make decisions based on the result. We verify the effectiveness of our model by analyzing a real purchase history data of a Japanese major retail company.